Spherical Imaging in Omnidirectional Camera Networks

We propose in this chapter to consider the emerging framework of networks of omnidirectional cameras. We first describe how omnidirectional images captured by different types of mirrors or lenses can be uniquely mapped to spherical images. Spherical imaging can then be used for calibration, scene analysis or distributed processing in omni-directional camera networks. The chapter then presents calibration methods that are specific to omnidirectional cameras. We observe the multi-view geometry framework with an omnivision perspective by re-formulating the epipolar geometry constraint for spherical projective imaging. In particular, we describe depth and disparity estimation in networks of omnidirectional cameras. Finally, we discuss the application of sparse approximations methods to spherical images, and we show how geometric representations can be used for distributed coding or scene understanding applications.